Thread: An AI Loyal to Its Owner Can Still Be Terrible for Everyone Else

imjustnewatai · x · 2026-09-13

A substantive thread pushing back on popular framings about "human control" and falling behind in the AI race:

An AI can be perfectly aligned with its owner and still harm everyone else — faithfully pursuing corporate profit or state advantage doesn't protect outsiders. "Human control" dodges the key question: which humans, and what stops abuse?
Being too far behind to win differs from being too far behind to cause serious damage; weaker systems can still cross dangerous capability thresholds.
If countries believe losing means permanently losing independence, even sensible leaders will take bigger risks — a single-winner narrative makes the race more reckless.
The author doubts China would accept permanent American AI dominance: weights already copied worldwide can't be recalled, and an unchallengeable private AI monopoly raises the same trust problem about who controls it.

More from AGI Musings